cocoon-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sylvain Wallez <sylvain.wal...@anyware-tech.com>
Subject Re: Flow result storage : let's fix it before its too late !
Date Tue, 04 Mar 2003 11:06:44 GMT
Stefano Mazzocchi wrote:

> Sylvain Wallez wrote:
>
>>>> So why don't we use the ObjectModel to pass the flow information ? 
>>>> It already contains the request and the response, and this seems 
>>>> the natural place for flow values. And more : if we consider 
>>>> components such as FlowVelocityGenerator, why would we want to 
>>>> publish only flow data and not elements of the object model ? If 
>>>> everything was in the object model, we could have some generic 
>>>> publication code that would publish all that is in the object 
>>>> model, regardless of what it actually is, and thus no need for a 
>>>> second VelocityGenerator.
>>>
>>>
>>> Well, actually I don't agree with you now. The flow script 
>>> establishes a contract with the presentation layer determined by the 
>>> object it passes to sendPage*() or the object it uses for the 
>>> XMLForm model. This, and only this, should be supplied to the 
>>> presentation layer. The presentation layer should not have access to 
>>> the request, session, etc, unless references to these are explicitly 
>>> passed by the flow script. 
>>
>
> Sorry, Sylvain, but I completely agree with Chris here.
>
>> I don't agree here : the flow is about business logic, and the 
>> presentation layer is about... presentation, which may often depend 
>> on information available in the request that aren't of any interest 
>> to the business logic. 
>
>
> nononononon
>
> The flow is about 'flow', it's about procedurally describing FSM 
> transitions. it's totally incorrect to state that 'flow := business 
> logic' because it is true that business logic influences the flow, but 
> the other way around it's not true.


Agree. My wording was too much restrictive.

> We don't use MVC, because MVC is too simple and mixes too many 
> concerns in the 'controller/model' realms. But at least clears the 
> fact that the 'view' is totally separated.
>
> if you start giving access from the view to the model, you make 
> concerns overlap because part of the flow-logic gets connected to the 
> 'view'.
>
>> To name a few, we have the target language, the user-agent, the host 
>> name (which can decide of the skin in case of virtual host), etc.
>
>
> The flow doesn't stop you from using ObjectModel-aware pipeline 
> components, but as far as generating the 'view' of a controller-driven 
> state, it's the controller that should drive the process. The 
> controller, in our case, it's the flow layer.
>
> Example: suppose that you want to display a request parameter in a XSP 
> that was triggered by a flow call. You could do it in two ways:
>
> 1) using accessing the request directly from the XSP
>
> 2) accessing a parameter given by the flow, which accessed the request 
> parameter.
>
> Which one is better?
>
> The second.
>
> Why? because the information on 'what' request parameter to use, was 
> probably dictated by the flow itself. So, the contract created by the 
> request parameter is only included in the flow layer and doesn't 
> crosscut the view part.
>
> The view has a totally different contract with the controller, which 
> is the name of the parameter passed by the flow.
>
> The better defined and well isolated the contracts, the cleaner and 
> easier SoC-oriented-programming becomes, reducing maintenance costs in 
> the long run by *orders* of magnitude. 


While this sounds good from a theoretical and "cleanliness" point of 
view, we have to consider that some information available in the object 
model is of no use for state transition, but is needed only for proper 
work of the view. So my question is "why should the flow bother about 
these" ?

With the flow bean dict becoming the _only_ source of information for 
the view, we are defining a strong and unique application-wide contract. 
But the problem with this unique contract is that it must be fed on one 
end to be consumed on the other end. This a push-driven model where the 
flow has to put in the dict each and every data consumed by the view, 
even if that data never influences neither state transition nor business 
logic. Enhydra's XMLC works that way and I find it bad.

 From a maintainance point of view, this means that the view *and* the 
flow have both to be modified in order to provide the view with some 
data the flow doesn't care about. So I'm not sure there's a real benefit 
here.

Having the object model available in the view allows a mixed push/pull 
model where the flow pushes the data it is responsible for and the view 
pulls data that is not the flow concern.

>> This is also a wonder I have about the current way the flow is 
>> perceived : by filling an important hole in Cocoon, it seems to me 
>> the flow is now surpassing its initial role. The flow isn't 
>> responsible for _all_ data that drive the system output, but only 
>> (and this should not be taken as a pejorative restriction) of the 
>> business data.
>
>
> Do you still feel the same after the context I've given you above?


Yep.

>>>> Thoughts ? What's the reason for using the Environment ?
>>>>
>>>> Also, I prefer longer but more explicit names than "$this" to 
>>>> access the flow bean : "$flowDict" is not much longer but way more 
>>>> explicit in the 
>>>
>>>
>>> Actually, under normal circumstances you never need to use $this. 
>>> Its only reason for existence is to provide a means of 
>>> disambiguating the built-in "continuation" property, in case the 
>>> bean also has a property called "continuation":
>>>
>>> $continuation - the continuation object
>>> $this.continuation - a property named "continuation"
>>
>>
>> The problem is not about "$this" but more about populating the flow 
>> data as toplevel variables. Explicitely naming the flow through a 
>> "$flowData" or something like that makes it more understandable about 
>> where this data comes from and where the user can know more about it.
>
>
> My vision is that people should be suggested to see the flow as *THE* 
> glue between views. In short, the natural programmatic side of the 
> sitemap, which is a role that now actions have and took because of no 
> real alternative.
>
> As much as we don't specify
>
>  <act ...>
>   <generate src="{action:whatever}"/>
>  </act>
>
> I really don't see why we should 'namespace' the flow variables more 
> either.
>
> But maybe there is something in your reasoning that I'm missing.


Yep. Basically, I feel the contract imposed by the flow to other 
components to be too strong and wanting to rule too much of the system.

Sylvain

-- 
Sylvain Wallez                                  Anyware Technologies
http://www.apache.org/~sylvain           http://www.anyware-tech.com
{ XML, Java, Cocoon, OpenSource }*{ Training, Consulting, Projects }



Mime
View raw message